User shares optimal settings for Qwen 3.6 27B on 48GB VRAM

User shares llama.cpp settings for running Qwen 3.6 27B (Q8_0) across dual RTX 4090 and RTX 3090 with tensor split, using 48GB total VRAM. The post includes specific configuration details from the community.